Support proc-macro lib type

[DenisGorbachev]

Hi, thank you for the crate!

I've implemented support for proc-macro lib type (see tests/pass/proc-macro). Currently, this is done in a backwards-compatible way: there is a new option --target-name that allows to specify the target name explicitly. The target name can be specified for a target of any type (including proc-macro).

Would you like to merge this PR?

[msrd0]

I'm not sure I understand the need for this change. Procedural macros are, afaik, just some special sort of library, and should be covered under library targets. Further, a package/crate/whateveryoucallit may only contain one library, so there is no need to distinguish between proc-macro libraries and "normal" libraries.

I ran the test case from this PR against the main branch and the test was successful.

Can you please elaborate why this change is necessary? What is the problem with the way proc-macro libraries are currently handled?

[DenisGorbachev]

Actually, "proc-macro" is a separate kind. The check target.is_lib() translates to target.is_kind("lib"), which doesn't match "proc-macro". So if the package has only a single target with kind = "proc-macro", then cargo-doc2readme can't find a suitable target (because it looks for kind = "lib" or kind = "bin").

Also, the test doesn't pass on my machine.

git checkout feature/target-proc-macro
cargo metadata --format-version 1 --manifest-path tests/pass/proc-macro/Cargo.toml | jq -r '.packages[0].targets[0] | {name, kind}' 
# {
#   "name": "proc_macro",
#   "kind": [
#     "proc-macro"
#   ]
# }
cargo doc2readme --out - --manifest-path tests/pass/proc-macro/Cargo.toml                   
# Error: Failed to find a library or binary target

[msrd0]

Oh I see, you marked the test case nightly-only, and I didn't run it with a nightly compiler :sweat_smile:

[msrd0]

I'm still not sure I support the approach you took here. From my perspective, there can be at most 1 library (proc-macro or not) per package, and I consider proc-macro to be a special case of libraries. So I would be extremely surprised if proc-macro was a separate type. Do you think it would make sense to keep the lib/bin separation but attempt to find a proc-macro target if a normal library target was not found?

[DenisGorbachev]

proc-macro is a separate kind, in fact :) Per cargo metadata:

git checkout feature/target-proc-macro
cargo metadata --format-version 1 --manifest-path tests/pass/proc-macro/Cargo.toml | jq -r '.packages[0].targets[0] | {name, kind}' 
# {
#   "name": "proc_macro",
#   "kind": [
#     "proc-macro"
#   ]
# }

For comparison, here's cargo metadata output for a regular lib:

cargo metadata --format-version 1 --manifest-path tests/pass/fn-link/Cargo.toml | jq -r '.packages[0].targets[0] | {name, kind}'
# {
#   "name": "fn_link",
#   "kind": [
#     "lib"
#   ]
# }

So the kind array contains either proc-macro for proc macro libs, or lib for regular libs. I know it's confusing, but that's how cargo metadata reports it.

Because of that, the check for target.is_lib() fails for proc-macro libs. Hence the PR that fixes it.